Niraj K. Jha

Shikhar Tuli, Niraj K. Jha

Self-attention-based transformer models have achieved tremendous success in the domain of natural language processing. Despite their efficacy, accelerating the transformer is challenging due to its quadratic computational complexity and large activation sizes. Existing transformer accelerators attempt to prune its tokens to reduce memory access, albeit with high compute overheads. Moreover, previous works directly operate on large matrices involved in the attention operation, which limits hardware utilization. In order to address these challenges, this work proposes a novel dynamic inference scheme, DynaTran, which prunes activations at runtime with low overhead, substantially reducing the number of ineffectual operations. This improves the throughput of transformer inference. We further propose tiling the matrices in transformer operations along with diverse dataflows to improve data reuse, thus enabling higher energy efficiency. To effectively implement these methods, we propose AccelTran, a novel accelerator architecture for transformers. Extensive experiments with different models and benchmarks demonstrate that DynaTran achieves higher accuracy than the state-of-the-art top-k hardware-aware pruning strategy while attaining up to 1.2$\times$ higher sparsity. One of our proposed accelerators, AccelTran-Edge, achieves 330K$\times$ higher throughput with 93K$\times$ lower energy requirement when compared to a Raspberry Pi device. On the other hand, AccelTran-Server achieves 5.73$\times$ higher throughput and 3.69$\times$ lower energy consumption compared to the state-of-the-art transformer co-processor, Energon. The simulation source code is available at https://github.com/jha-lab/acceltran.

Yihao Liang, Ze Wang, Hao Chen et al.

Autoregressive large language models achieve strong results on many benchmarks, but decoding remains fundamentally latency-limited by sequential dependence on previously generated tokens. Diffusion language models (DLMs) promise parallel generation but suffer from a fundamental static-to-dynamic misalignment: Training optimizes local transitions under fixed schedules, whereas efficient inference requires adaptive "long-jump" refinements through unseen states. Our goal is to enable highly parallel decoding for DLMs with low number of function evaluations while preserving generation quality. To achieve this, we propose CD4LM, a framework that decouples training from inference via Discrete-Space Consistency Distillation (DSCD) and Confidence-Adaptive Decoding (CAD). Unlike standard objectives, DSCD trains a student to be trajectory-invariant, mapping diverse noisy states directly to the clean distribution. This intrinsic robustness enables CAD to dynamically allocate compute resources based on token confidence, aggressively skipping steps without the quality collapse typical of heuristic acceleration. On GSM8K, CD4LM matches the LLaDA baseline with a 5.18x wall-clock speedup; across code and math benchmarks, it strictly dominates the accuracy-efficiency Pareto frontier, achieving a 3.62x mean speedup while improving average accuracy. Code is available at https://github.com/yihao-liang/CDLM

Chang Yue, Niraj K. Jha

In supervised machine learning, use of correct labels is extremely important to ensure high accuracy. Unfortunately, most datasets contain corrupted labels. Machine learning models trained on such datasets do not generalize well. Thus, detecting their label errors can significantly increase their efficacy. We propose a novel framework, called CTRL (Clustering TRaining Losses for label error detection), to detect label errors in multi-class datasets. It detects label errors in two steps based on the observation that models learn clean and noisy labels in different ways. First, we train a neural network using the noisy training dataset and obtain the loss curve for each sample. Then, we apply clustering algorithms to the training losses to group samples into two categories: cleanly-labeled and noisily-labeled. After label error detection, we remove samples with noisy labels and retrain the model. Our experimental results demonstrate state-of-the-art error detection accuracy on both image (CIFAR-10 and CIFAR-100) and tabular datasets under simulated noise. We also use a theoretical analysis to provide insights into why CTRL performs so well.

Shikhar Tuli, Chia-Hao Li, Ritvik Sharma et al.

Recently, automated co-design of machine learning (ML) models and accelerator architectures has attracted significant attention from both the industry and academia. However, most co-design frameworks either explore a limited search space or employ suboptimal exploration techniques for simultaneous design decision investigations of the ML model and the accelerator. Furthermore, training the ML model and simulating the accelerator performance is computationally expensive. To address these limitations, this work proposes a novel neural architecture and hardware accelerator co-design framework, called CODEBench. It is composed of two new benchmarking sub-frameworks, CNNBench and AccelBench, which explore expanded design spaces of convolutional neural networks (CNNs) and CNN accelerators. CNNBench leverages an advanced search technique, BOSHNAS, to efficiently train a neural heteroscedastic surrogate model to converge to an optimal CNN architecture by employing second-order gradients. AccelBench performs cycle-accurate simulations for a diverse set of accelerator architectures in a vast design space. With the proposed co-design method, called BOSHCODE, our best CNN-accelerator pair achieves 1.4% higher accuracy on the CIFAR-10 dataset compared to the state-of-the-art pair, while enabling 59.1% lower latency and 60.8% lower energy consumption. On the ImageNet dataset, it achieves 3.7% higher Top1 accuracy at 43.8% lower latency and 11.2% lower energy consumption. CODEBench outperforms the state-of-the-art framework, i.e., Auto-NBA, by achieving 1.5% higher accuracy and 34.7x higher throughput, while enabling 11.0x lower energy-delay product (EDP) and 4.0x lower chip area on CIFAR-10.

Shikhar Tuli, Bhishma Dedhia, Shreshth Tuli et al.

The existence of a plethora of language models makes the problem of selecting the best one for a custom task challenging. Most state-of-the-art methods leverage transformer-based models (e.g., BERT) or their variants. Training such models and exploring their hyperparameter space, however, is computationally expensive. Prior work proposes several neural architecture search (NAS) methods that employ performance predictors (e.g., surrogate models) to address this issue; however, analysis has been limited to homogeneous models that use fixed dimensionality throughout the network. This leads to sub-optimal architectures. To address this limitation, we propose a suite of heterogeneous and flexible models, namely FlexiBERT, that have varied encoder layers with a diverse set of possible operations and different hidden dimensions. For better-posed surrogate modeling in this expanded design space, we propose a new graph-similarity-based embedding scheme. We also propose a novel NAS policy, called BOSHNAS, that leverages this new scheme, Bayesian modeling, and second-order optimization, to quickly train and use a neural surrogate model to converge to the optimal architecture. A comprehensive set of experiments shows that the proposed policy, when applied to the FlexiBERT design space, pushes the performance frontier upwards compared to traditional models. FlexiBERT-Mini, one of our proposed models, has 3% fewer parameters than BERT-Mini and achieves 8.9% higher GLUE score. A FlexiBERT model with equivalent performance as the best homogeneous model achieves 2.6x smaller size. FlexiBERT-Large, another proposed model, achieves state-of-the-art results, outperforming the baseline models by at least 5.7% on the GLUE benchmark.

Shikhar Tuli, Niraj K. Jha

Automated co-design of machine learning models and evaluation hardware is critical for efficiently deploying such models at scale. Despite the state-of-the-art performance of transformer models, they are not yet ready for execution on resource-constrained hardware platforms. High memory requirements and low parallelizability of the transformer architecture exacerbate this problem. Recently-proposed accelerators attempt to optimize the throughput and energy consumption of transformer models. However, such works are either limited to a one-sided search of the model architecture or a restricted set of off-the-shelf devices. Furthermore, previous works only accelerate model inference and not training, which incurs substantially higher memory and compute resources, making the problem even more challenging. To address these limitations, this work proposes a dynamic training framework, called DynaProp, that speeds up the training process and reduces memory consumption. DynaProp is a low-overhead pruning method that prunes activations and gradients at runtime. To effectively execute this method on hardware for a diverse set of transformer architectures, we propose ELECTOR, a framework that simulates transformer inference and training on a design space of accelerators. We use this simulator in conjunction with the proposed co-design technique, called TransCODE, to obtain the best-performing models with high accuracy on the given task and minimize latency, energy consumption, and chip area. The obtained transformer-accelerator pair achieves 0.3% higher accuracy than the state-of-the-art pair while incurring 5.2$\times$ lower latency and 3.0$\times$ lower energy consumption.

Bhishma Dedhia, Roshini Balasubramanian, Niraj K. Jha

The Synthetic Control method has pioneered a class of powerful data-driven techniques to estimate the counterfactual reality of a unit from donor units. At its core, the technique involves a linear model fitted on the pre-intervention period that combines donor outcomes to yield the counterfactual. However, linearly combining spatial information at each time instance using time-agnostic weights fails to capture important inter-unit and intra-unit temporal contexts and complex nonlinear dynamics of real data. We instead propose an approach to use local spatiotemporal information before the onset of the intervention as a promising way to estimate the counterfactual sequence. To this end, we suggest a Transformer model that leverages particular positional embeddings, a modified decoder attention mask, and a novel pre-training task to perform spatiotemporal sequence-to-sequence modeling. Our experiments on synthetic data demonstrate the efficacy of our method in the typical small donor pool setting and its robustness against noise. We also generate actionable healthcare insights at the population and patient levels by simulating a state-wide public health policy to evaluate its effectiveness, an in silico trial for asthma medications to support randomized controlled trials, and a medical intervention for patients with Friedreich's ataxia to improve clinical decision-making and promote personalized therapy.

Shikhar Tuli, Niraj K. Jha

Automated design of efficient transformer models has recently attracted significant attention from industry and academia. However, most works only focus on certain metrics while searching for the best-performing transformer architecture. Furthermore, running traditional, complex, and large transformer models on low-compute edge platforms is a challenging problem. In this work, we propose a framework, called ProTran, to profile the hardware performance measures for a design space of transformer architectures and a diverse set of edge devices. We use this profiler in conjunction with the proposed co-design technique to obtain the best-performing models that have high accuracy on the given task and minimize latency, energy consumption, and peak power draw to enable edge deployment. We refer to our framework for co-optimizing accuracy and hardware performance measures as EdgeTran. It searches for the best transformer model and edge device pair. Finally, we propose GPTran, a multi-stage block-level grow-and-prune post-processing step that further improves accuracy in a hardware-aware manner. The obtained transformer model is 2.8$\times$ smaller and has a 0.8% higher GLUE score than the baseline (BERT-Base). Inference with it on the selected edge device enables 15.0% lower latency, 10.0$\times$ lower energy, and 10.8$\times$ lower peak power draw compared to an off-the-shelf GPU.

Chang Yue, Niraj K. Jha

The ubiquity of neural networks (NNs) in real-world applications, from healthcare to natural language processing, underscores their immense utility in capturing complex relationships within high-dimensional data. However, NNs come with notable disadvantages, such as their "black-box" nature, which hampers interpretability, as well as their tendency to overfit the training data. We introduce a novel method for learning interpretable differentiable logic networks (DLNs) that are architectures that employ multiple layers of binary logic operators. We train these networks by softening and differentiating their discrete components, e.g., through binarization of inputs, binary logic operations, and connections between neurons. This approach enables the use of gradient-based learning methods. Experimental results on twenty classification tasks indicate that differentiable logic networks can achieve accuracies comparable to or exceeding that of traditional NNs. Equally importantly, these networks offer the advantage of interpretability. Moreover, their relatively simple structure results in the number of logic gate-level operations during inference being up to a thousand times smaller than NNs, making them suitable for deployment on edge devices.

Ian Henriques, Lynda Elhassar, Sarvesh Relekar et al.

The global rise in type 2 diabetes underscores the need for scalable and cost-effective screening methods. Current diagnosis requires biochemical assays, which are invasive and costly. Advances in consumer wearables have enabled early explorations of machine learning-based disease detection, but prior studies were limited to controlled settings. We present SweetDeep, a compact neural network trained on physiological and demographic data from 285 (diabetic and non-diabetic) participants in the EU and MENA regions, collected using Samsung Galaxy Watch 7 devices in free-living conditions over six days. Each participant contributed multiple 2-minute sensor recordings per day, totaling approximately 20 recordings per individual. Despite comprising fewer than 3,000 parameters, SweetDeep achieves 82.5% patient-level accuracy (82.1% macro-F1, 79.7% sensitivity, 84.6% specificity) under three-fold cross-validation, with an expected calibration error of 5.5%. Allowing the model to abstain on less than 10% of low-confidence patient predictions yields an accuracy of 84.5% on the remaining patients. These findings demonstrate that combining engineered features with lightweight architectures can support accurate, rapid, and generalizable detection of type 2 diabetes in real-world wearable settings.

Shikhar Tuli, Niraj K. Jha

Researchers constantly strive to explore larger and more complex search spaces in various scientific studies and physical experiments. However, such investigations often involve sophisticated simulators or time-consuming experiments that make exploring and observing new design samples challenging. Previous works that target such applications are typically sample-inefficient and restricted to vector search spaces. To address these limitations, this work proposes a constrained multi-objective optimization (MOO) framework, called BREATHE, that searches not only traditional vector-based design spaces but also graph-based design spaces to obtain best-performing graphs. It leverages second-order gradients and actively trains a heteroscedastic surrogate model for sample-efficient optimization. In a single-objective vector optimization application, it leads to 64.1% higher performance than the next-best baseline, random forest regression. In graph-based search, BREATHE outperforms the next-best baseline, i.e., a graphical version of Gaussian-process-based Bayesian optimization, with up to 64.9% higher performance. In a MOO task, it achieves up to 21.9$\times$ higher hypervolume than the state-of-the-art method, multi-objective Bayesian optimization (MOBOpt). BREATHE also outperforms the baseline methods on most standard MOO benchmark applications.

Chia-Hao Li, Niraj K. Jha

Wearable medical sensors (WMSs) are revolutionizing smart healthcare by enabling continuous, real-time monitoring of user physiological signals, especially in the field of consumer healthcare. The integration of WMSs and modern machine learning (ML) enables unprecedented solutions to efficient early-stage disease detection. Despite the success of Transformers in various fields, their application to sensitive domains, such as smart healthcare, remains underexplored due to limited data accessibility and privacy concerns. To bridge the gap between Transformer-based foundation models and WMS-based disease detection, we propose COMFORT, a continual fine-tuning framework for foundation models targeted at consumer healthcare. COMFORT introduces a novel approach for pre-training a Transformer-based foundation model on a large dataset of physiological signals exclusively collected from healthy individuals with commercially available WMSs. We adopt a masked data modeling (MDM) objective to pre-train this health foundation model. We then fine-tune the model using various parameter-efficient fine-tuning (PEFT) methods, such as low-rank adaptation (LoRA) and its variants, to adapt it to various downstream disease detection tasks that rely on WMS data. In addition, COMFORT continually stores the low-rank decomposition matrices obtained from the PEFT algorithms to construct a library for multi-disease detection. The COMFORT library enables scalable and memory-efficient disease detection on edge devices. Our experimental results demonstrate that COMFORT achieves highly competitive performance while reducing memory overhead by up to 52% relative to conventional methods. Thus, COMFORT paves the way for personalized and proactive solutions to efficient and effective early-stage disease detection for consumer healthcare.

Yuval Kansal, Niraj K. Jha

Large language models have achieved near-expert performance in structured reasoning domains like mathematics and programming, yet their ability to perform compositional multi-hop reasoning in specialized scientific fields remains limited. We propose a bottom-up learning paradigm in which models are grounded in axiomatic domain facts and compose them to solve complex, unseen tasks. To this end, we present a post-training pipeline, based on a combination of supervised fine-tuning and reinforcement learning (RL), in which knowledge graphs act as implicit reward models. By deriving novel reward signals from knowledge graph paths, we provide verifiable, scalable, and grounded supervision that encourages models to compose intermediate axioms rather than optimize only final answers during RL. We validate this approach in the medical domain, training a 14B model on short-hop reasoning paths (1-3 hops) and evaluating its zero-shot generalization to complex multi-hop queries (4-5 hops). Our experiments show that path-derived rewards act as a "compositional bridge", enabling our model to significantly outperform much larger models and frontier systems like GPT-5.2 and Gemini 3 Pro, on the most difficult reasoning tasks. Furthermore, we demonstrate the robustness of our approach to adversarial perturbations against option-shuffling stress tests. This work suggests that grounding the reasoning process in structured knowledge is a scalable and efficient path toward intelligent reasoning.

Hongxu Yin, Pavlo Molchanov, Zhizhong Li et al.

We introduce DeepInversion, a new method for synthesizing images from the image distribution used to train a deep neural network. We 'invert' a trained network (teacher) to synthesize class-conditional input images starting from random noise, without using any additional information about the training dataset. Keeping the teacher fixed, our method optimizes the input while regularizing the distribution of intermediate feature maps using information stored in the batch normalization layers of the teacher. Further, we improve the diversity of synthesized images using Adaptive DeepInversion, which maximizes the Jensen-Shannon divergence between the teacher and student network logits. The resulting synthesized images from networks trained on the CIFAR-10 and ImageNet datasets demonstrate high fidelity and degree of realism, and help enable a new breed of data-free applications - ones that do not require any real images or labeled data. We demonstrate the applicability of our proposed method to three tasks of immense practical importance -- (i) data-free network pruning, (ii) data-free knowledge transfer, and (iii) data-free continual learning. Code is available at https://github.com/NVlabs/DeepInversion

Hongjie Wang, Difan Liu, Yan Kang et al.

Diffusion Models (DMs) have exhibited superior performance in generating high-quality and diverse images. However, this exceptional performance comes at the cost of expensive architectural design, particularly due to the attention module heavily used in leading models. Existing works mainly adopt a retraining process to enhance DM efficiency. This is computationally expensive and not very scalable. To this end, we introduce the Attention-driven Training-free Efficient Diffusion Model (AT-EDM) framework that leverages attention maps to perform run-time pruning of redundant tokens, without the need for any retraining. Specifically, for single-denoising-step pruning, we develop a novel ranking algorithm, Generalized Weighted Page Rank (G-WPR), to identify redundant tokens, and a similarity-based recovery method to restore tokens for the convolution operation. In addition, we propose a Denoising-Steps-Aware Pruning (DSAP) approach to adjust the pruning budget across different denoising timesteps for better generation quality. Extensive evaluations show that AT-EDM performs favorably against prior art in terms of efficiency (e.g., 38.8% FLOPs saving and up to 1.53x speed-up over Stable Diffusion XL) while maintaining nearly the same FID and CLIP scores as the full model. Project webpage: https://atedm.github.io.

Hongjie Wang, Chih-Yao Ma, Yen-Cheng Liu et al.

Text-to-video generation enhances content creation but is highly computationally intensive: The computational cost of Diffusion Transformers (DiTs) scales quadratically in the number of pixels. This makes minute-length video generation extremely expensive, limiting most existing models to generating videos of only 10-20 seconds length. We propose a Linear-complexity text-to-video Generation (LinGen) framework whose cost scales linearly in the number of pixels. For the first time, LinGen enables high-resolution minute-length video generation on a single GPU without compromising quality. It replaces the computationally-dominant and quadratic-complexity block, self-attention, with a linear-complexity block called MATE, which consists of an MA-branch and a TE-branch. The MA-branch targets short-to-long-range correlations, combining a bidirectional Mamba2 block with our token rearrangement method, Rotary Major Scan, and our review tokens developed for long video generation. The TE-branch is a novel TEmporal Swin Attention block that focuses on temporal correlations between adjacent tokens and medium-range tokens. The MATE block addresses the adjacency preservation issue of Mamba and improves the consistency of generated videos significantly. Experimental results show that LinGen outperforms DiT (with a 75.6% win rate) in video quality with up to 15$\times$ (11.5$\times$) FLOPs (latency) reduction. Furthermore, both automatic metrics and human evaluation demonstrate our LinGen-4B yields comparable video quality to state-of-the-art models (with a 50.5%, 52.1%, 49.1% win rate with respect to Gen-3, LumaLabs, and Kling, respectively). This paves the way to hour-length movie generation and real-time interactive video generation. We provide 68s video generation results and more examples in our project website: https://lineargen.github.io/.

Bhishma Dedhia, Yuval Kansal, Niraj K. Jha

Language models traditionally used for cross-domain generalization have recently demonstrated task-specific reasoning. However, their top-down training approach on general corpora is insufficient for acquiring abstractions needed for deep domain expertise. This may require a bottom-up approach that acquires expertise by learning to compose simple domain concepts into more complex ones. A knowledge graph (KG) provides this compositional structure, where domain primitives are represented as head-relation-tail edges and their paths encode higher-level concepts. We present a task generation pipeline that synthesizes tasks directly from KG primitives, enabling models to acquire and compose them for reasoning. We fine-tune language models on the resultant KG-grounded curriculum to demonstrate domain-specific superintelligence. While broadly applicable, we validate our approach in medicine, where reliable KGs exist. Using a medical KG, we curate 24,000 reasoning tasks paired with thinking traces derived from diverse medical primitives. We fine-tune the QwQ-32B model on this curriculum to obtain QwQ-Med-3 that takes a step towards medical superintelligence. We also introduce ICD-Bench, an evaluation suite to quantify reasoning abilities across 15 medical domains. Our experiments demonstrate that QwQ-Med-3 significantly outperforms state-of-the-art reasoning models on ICD-Bench categories. Further analysis reveals that QwQ-Med-3 utilizes acquired primitives to widen the performance gap on the hardest tasks of ICD-Bench. Finally, evaluation on medical question-answer benchmarks shows that QwQ-Med-3 transfers acquired expertise to enhance the base model's performance. While the industry's approach to artificial general intelligence (AGI) emphasizes broad expertise, we envision a future in which AGI emerges from the composable interaction of efficient domain-specific superintelligent agents.

Shikhar Tuli, Chi-Heng Lin, Yen-Chang Hsu et al.

Traditional language models operate autoregressively, i.e., they predict one token at a time. Rapid explosion in model sizes has resulted in high inference times. In this work, we propose DynaMo, a suite of multi-token prediction language models that reduce net inference times. Our models $\textit{dynamically}$ predict multiple tokens based on their confidence in the predicted joint probability distribution. We propose a lightweight technique to train these models, leveraging the weights of traditional autoregressive counterparts. Moreover, we propose novel ways to enhance the estimated joint probability to improve text generation quality, namely co-occurrence weighted masking and adaptive thresholding. We also propose systematic qualitative and quantitative methods to rigorously test the quality of generated text for non-autoregressive generation. One of the models in our suite, DynaMo-7.3B-T3, achieves same-quality generated text as the baseline (Pythia-6.9B) while achieving 2.57$\times$ speed-up with only 5.87% and 2.67% parameter and training time overheads, respectively.

Chang Yue, Niraj K. Jha

Neural networks (NNs) achieve outstanding performance in many domains; however, their decision processes are often opaque and their inference can be computationally expensive in resource-constrained environments. We recently proposed Differentiable Logic Networks (DLNs) to address these issues for tabular classification based on relaxing discrete logic into a differentiable form, thereby enabling gradient-based learning of networks built from binary logic operations. DLNs offer interpretable reasoning and substantially lower inference cost. We extend the DLN framework to supervised tabular regression. Specifically, we redesign the final output layer to support continuous targets and unify the original two-phase training procedure into a single differentiable stage. We evaluate the resulting model on 15 public regression benchmarks, comparing it with modern neural networks and classical regression baselines. Regression DLNs match or exceed baseline accuracy while preserving interpretability and fast inference. Our results show that DLNs are a viable, cost-effective alternative for regression tasks, especially where model transparency and computational efficiency are important.

Bhishma Dedhia, Niraj K. Jha

Several accounts of human cognition posit that our intelligence is rooted in our ability to form abstract composable concepts, ground them in our environment, and reason over these grounded entities. This trifecta of human thought has remained elusive in modern intelligent machines. In this work, we investigate whether slot representations extracted from visual scenes serve as appropriate compositional abstractions for grounding and reasoning. We present the Neural Slot Interpreter (NSI), which learns to ground object semantics in slots. At the core of NSI is a nested schema that uses simple syntax rules to organize the object semantics of a scene into object-centric schema primitives. Then, the NSI metric learns to ground primitives into slots through a structured contrastive learning objective that reasons over the intermodal alignment. Experiments with a bi-modal object-property and scene retrieval task demonstrate the grounding efficacy and interpretability of correspondences learned by NSI. From a scene representation standpoint, we find that emergent NSI slots that move beyond the image grid by binding to spatial objects facilitate improved visual grounding compared to conventional bounding-box-based approaches. From a data efficiency standpoint, we empirically validate that NSI learns more generalizable representations from a fixed amount of annotation data than the traditional approach. We also show that the grounded slots surpass unsupervised slots in real-world object discovery and scale with scene complexity. Finally, we investigate the downstream efficacy of the grounded slots. Vision Transformers trained on grounding-aware NSI tokenizers using as few as ten tokens outperform patch-based tokens on challenging few-shot classification tasks.

Hongjie Wang, Niraj K. Jha

Modern Vision-Language Models (VLMs) achieve impressive performance but are limited by the quadratic complexity of self-attention, which prevents their deployment on edge devices and makes their understanding of high-resolution images and long-context videos prohibitively expensive. To address this challenge, we introduce LinMU (Linear-complexity Multimodal Understanding), a VLM design that achieves linear complexity without using any quadratic-complexity modules while maintaining the performance of global-attention-based VLMs. LinMU replaces every self-attention layer in the VLM with the M-MATE block: a dual-branch module that combines a bidirectional state-space model for global context (Flex-MA branch) with localized Swin-style window attention (Local-Swin branch) for adjacent correlations. To transform a pre-trained VLM into the LinMU architecture, we propose a three-stage distillation framework that (i) initializes both branches with self-attention weights and trains the Flex-MA branch alone, (ii) unfreezes the Local-Swin branch and fine-tunes it jointly with the Flex-MA branch, and (iii) unfreezes the remaining blocks and fine-tunes them using LoRA adapters, while regressing on hidden states and token-level logits of the frozen VLM teacher. On MMMU, TextVQA, LongVideoBench, Video-MME, and other benchmarks, LinMU matches the performance of teacher models, yet reduces Time-To-First-Token (TTFT) by up to 2.7$\times$ and improves token throughput by up to 9.0$\times$ on minute-length videos. Ablations confirm the importance of each distillation stage and the necessity of the two branches of the M-MATE block. The proposed framework demonstrates that state-of-the-art multimodal reasoning can be achieved without quadratic attention, thus opening up avenues for long-context VLMs that can deal with high-resolution images and long videos.

Margarita Belova, Jiaxin Xiao, Shikhar Tuli et al.

Researchers have pursued neurosymbolic artificial intelligence (AI) applications for nearly three decades because symbolic components provide abstraction while neural components provide generalization. Thus, a marriage of the two components can lead to rapid advancements in AI. Yet, the field has not realized this promise since most neurosymbolic AI frameworks fail to scale. In addition, the implicit representations and approximate reasoning of neural approaches limit interpretability and trust. Knowledge graphs (KGs), a gold-standard representation of explicit semantic knowledge, can address the symbolic side. However, automatically deriving reliable KGs from text corpora has remained an open problem. We address these challenges by introducing GraphMERT, a tiny graphical encoder-only model that distills high-quality KGs from unstructured text corpora and its own internal representations. GraphMERT and its equivalent KG form a modular neurosymbolic stack: neural learning of abstractions; symbolic KGs for verifiable reasoning. GraphMERT + KG is the first efficient and scalable neurosymbolic model to achieve state-of-the-art benchmark accuracy along with superior symbolic representations relative to baselines. Concretely, we target reliable domain-specific KGs that are both (1) factual (with provenance) and (2) valid (ontology-consistent relations with domain-appropriate semantics). When a large language model (LLM), e.g., Qwen3-32B, generates domain-specific KGs, it falls short on reliability due to prompt sensitivity, shallow domain expertise, and hallucinated relations. On text obtained from PubMed papers on diabetes, our 80M-parameter GraphMERT yields a KG with a 69.8% FActScore; a 32B-parameter baseline LLM yields a KG that achieves only 40.2% FActScore. The GraphMERT KG also attains a higher ValidityScore of 68.8%, versus 43.0% for the LLM baseline.

Sayeri Lala, Niraj K. Jha

Clinical randomized controlled trials (RCTs) collect hundreds of measurements spanning various metric types (e.g., laboratory tests, cognitive/motor assessments, etc.) across 100s-1000s of subjects to evaluate the effect of a treatment, but do so at the cost of significant trial expense. To reduce the number of measurements, trial protocols can be revised to remove metrics extraneous to the study's objective, but doing so requires additional human labor and limits the set of hypotheses that can be studied with the collected data. In contrast, a planned missing design (PMD) can reduce the amount of data collected without removing any metric by imputing the unsampled data. Standard PMDs randomly sample data to leverage statistical properties of imputation algorithms, but are ad hoc, hence suboptimal. Methods that learn PMDs produce more sample-efficient PMDs, but are not suitable for RCTs because they require ample prior data (150+ subjects) to model the data distribution. Therefore, we introduce a framework called Measurement EfficienT Randomized Controlled Trials using Transformers with Input MasKing (METRIK), which, for the first time, calculates a PMD specific to the RCT from a modest amount of prior data (e.g., 60 subjects). Specifically, METRIK models the PMD as a learnable input masking layer that is optimized with a state-of-the-art imputer based on the Transformer architecture. METRIK implements a novel sampling and selection algorithm to generate a PMD that satisfies the trial designer's objective, i.e., whether to maximize sampling efficiency or imputation performance for a given sampling budget. Evaluated across five real-world clinical RCT datasets, METRIK increases the sampling efficiency of and imputation performance under the generated PMD by leveraging correlations over time and across metrics, thereby removing the need to manually remove metrics from the RCT.

Linhui Huang, Sayeri Lala, Niraj K. Jha

Deep neural networks exhibit remarkable performance, yet their black-box nature limits their utility in fields like healthcare where interpretability is crucial. Existing explainability approaches often sacrifice accuracy and lack quantifiable measures of prediction uncertainty. In this study, we introduce Conformal Prediction for Interpretable Neural Networks (CONFINE), a versatile framework that generates prediction sets with statistically robust uncertainty estimates instead of point predictions to enhance model transparency and reliability. CONFINE not only provides example-based explanations and confidence estimates for individual predictions but also boosts accuracy by up to 3.6%. We define a new metric, correct efficiency, to evaluate the fraction of prediction sets that contain precisely the correct label and show that CONFINE achieves correct efficiency of up to 3.3% higher than the original accuracy, matching or exceeding prior methods. CONFINE's marginal and class-conditional coverages attest to its validity across tasks spanning medical image classification to language understanding. Being adaptable to any pre-trained classifier, CONFINE marks a significant advance towards transparent and trustworthy deep learning applications in critical domains.

Chia-Hao Li, Niraj K. Jha

We propose PAGE, a domain-incremental adaptation strategy with past-agnostic generative replay for smart healthcare. PAGE enables generative replay without the aid of any preserved data or information from prior domains. When adapting to a new domain, it exploits real data from the new distribution and the current model to generate synthetic data that retain the learned knowledge of previous domains. By replaying the synthetic data with the new real data during training, PAGE achieves a good balance between domain adaptation and knowledge retention. In addition, we incorporate an extended inductive conformal prediction (EICP) method into PAGE to produce a confidence score and a credibility value for each detection result. This makes the predictions interpretable and provides statistical guarantees for disease detection in smart healthcare applications. We demonstrate PAGE's effectiveness in domain-incremental disease detection with three distinct disease datasets collected from commercially available WMSs. PAGE achieves highly competitive performance against state-of-the-art with superior scalability, data privacy, and feasibility. Furthermore, PAGE can enable up to 75% reduction in clinical workload with the help of EICP.

Hongjie Wang, Bhishma Dedhia, Niraj K. Jha

Deployment of Transformer models on edge devices is becoming increasingly challenging due to the exponentially growing inference cost that scales quadratically with the number of tokens in the input sequence. Token pruning is an emerging solution to address this challenge due to its ease of deployment on various Transformer backbones. However, most token pruning methods require computationally expensive fine-tuning, which is undesirable in many edge deployment cases. In this work, we propose Zero-TPrune, the first zero-shot method that considers both the importance and similarity of tokens in performing token pruning. It leverages the attention graph of pre-trained Transformer models to produce an importance distribution for tokens via our proposed Weighted Page Rank (WPR) algorithm. This distribution further guides token partitioning for efficient similarity-based pruning. Due to the elimination of the fine-tuning overhead, Zero-TPrune can prune large models at negligible computational cost, switch between different pruning configurations at no computational cost, and perform hyperparameter tuning efficiently. We evaluate the performance of Zero-TPrune on vision tasks by applying it to various vision Transformer backbones and testing them on ImageNet. Without any fine-tuning, Zero-TPrune reduces the FLOPs cost of DeiT-S by 34.7% and improves its throughput by 45.3% with only 0.4% accuracy loss. Compared with state-of-the-art pruning methods that require fine-tuning, Zero-TPrune not only eliminates the need for fine-tuning after pruning but also does so with only 0.1% accuracy loss. Compared with state-of-the-art fine-tuning-free pruning methods, Zero-TPrune reduces accuracy loss by up to 49% with similar FLOPs budgets. Project webpage: https://jha-lab.github.io/zerotprune.

Bhishma Dedhia, Michael Chang, Jake C. Snell et al.

Large language models are few-shot learners that can solve diverse tasks from a handful of demonstrations. This implicit understanding of tasks suggests that the attention mechanisms over word tokens may play a role in analogical reasoning. In this work, we investigate whether analogical reasoning can enable in-context composition over composable elements of visual stimuli. First, we introduce a suite of three benchmarks to test the generalization properties of a visual in-context learner. We formalize the notion of an analogy-based in-context learner and use it to design a meta-learning framework called Im-Promptu. Whereas the requisite token granularity for language is well established, the appropriate compositional granularity for enabling in-context generalization in visual stimuli is usually unspecified. To this end, we use Im-Promptu to train multiple agents with different levels of compositionality, including vector representations, patch representations, and object slots. Our experiments reveal tradeoffs between extrapolation abilities and the degree of compositionality, with non-compositional representations extending learned composition rules to unseen domains but performing poorly on combinatorial tasks. Patch-based representations require patches to contain entire objects for robust extrapolation. At the same time, object-centric tokenizers coupled with a cross-attention module generate consistent and high-fidelity solutions, with these inductive biases being particularly crucial for compositional generalization. Lastly, we demonstrate a use case of Im-Promptu as an intuitive programming interface for image generation.

Chia-Hao Li, Niraj K. Jha

Modern advances in machine learning (ML) and wearable medical sensors (WMSs) in edge devices have enabled ML-driven disease detection for smart healthcare. Conventional ML-driven methods for disease detection rely on customizing individual models for each disease and its corresponding WMS data. However, such methods lack adaptability to distribution shifts and new task classification classes. In addition, they need to be rearchitected and retrained from scratch for each new disease. Moreover, installing multiple ML models in an edge device consumes excessive memory, drains the battery faster, and complicates the detection process. To address these challenges, we propose DOCTOR, a multi-disease detection continual learning (CL) framework based on WMSs. It employs a multi-headed deep neural network (DNN) and a replay-style CL algorithm. The CL algorithm enables the framework to continually learn new missions where different data distributions, classification classes, and disease detection tasks are introduced sequentially. It counteracts catastrophic forgetting with a data preservation method and a synthetic data generation (SDG) module. The data preservation method preserves the most informative subset of real training data from previous missions for exemplar replay. The SDG module models the probability distribution of the real training data and generates synthetic data for generative replay while retaining data privacy. The multi-headed DNN enables DOCTOR to detect multiple diseases simultaneously based on user WMS data. We demonstrate DOCTOR's efficacy in maintaining high disease classification accuracy with a single DNN model in various CL experiments. In complex scenarios, DOCTOR achieves 1.43 times better average test accuracy, 1.25 times better F1-score, and 0.41 higher backward transfer than the naive fine-tuning framework with a small model size of less than 350KB.

Sayeri Lala, Niraj K. Jha

The randomized controlled trial (RCT) is the gold standard for estimating the average treatment effect (ATE) of a medical intervention but requires 100s-1000s of subjects, making it expensive and difficult to implement. While a cross-over trial can reduce sample size requirements by measuring the treatment effect per individual, it is only applicable to chronic conditions and interventions whose effects dissipate rapidly. Another approach is to replace or augment data collected from an RCT with external data from prospective studies or prior RCTs, but it is vulnerable to confounders in the external or augmented data. We propose to simulate the cross-over trial to overcome its practical limitations while exploiting its strengths. We propose a novel framework, SECRETS, which, for the first time, estimates the individual treatment effect (ITE) per patient in the RCT study without using any external data by leveraging a state-of-the-art counterfactual estimation algorithm, called synthetic intervention. It also uses a new hypothesis testing strategy to determine whether the treatment has a clinically significant ATE based on the estimated ITEs. We show that SECRETS can improve the power of an RCT while maintaining comparable significance levels; in particular, on three real-world clinical RCTs (Phase-3 trials), SECRETS increases power over the baseline method by $\boldsymbol{6}$-$\boldsymbol{54\%}$ (average: 21.5%, standard deviation: 15.8%).

Tanujay Saha, Najwa Aaraj, Niraj K. Jha

The core network architecture of telecommunication systems has undergone a paradigm shift in the fifth-generation (5G)networks. 5G networks have transitioned to software-defined infrastructures, thereby reducing their dependence on hardware-based network functions. New technologies, like network function virtualization and software-defined networking, have been incorporated in the 5G core network (5GCN) architecture to enable this transition. This has resulted in significant improvements in efficiency, performance, and robustness of the networks. However, this has also made the core network more vulnerable, as software systems are generally easier to compromise than hardware systems. In this article, we present a comprehensive security analysis framework for the 5GCN. The novelty of this approach lies in the creation and analysis of attack graphs of the software-defined and virtualized 5GCN through machine learning. This analysis points to 119 novel possible exploits in the 5GCN. We demonstrate that these possible exploits of 5GCN vulnerabilities generate five novel attacks on the 5G Authentication and Key Agreement protocol. We combine the attacks at the network, protocol, and the application layers to generate complex attack vectors. In a case study, we use these attack vectors to find four novel security loopholes in WhatsApp running on a 5G network.

Jacob Brown, Tanujay Saha, Niraj K. Jha

Internet-of-Things (IoT) and cyber-physical systems (CPSs) may consist of thousands of devices connected in a complex network topology. The diversity and complexity of these components present an enormous attack surface, allowing an adversary to exploit security vulnerabilities of different devices to execute a potent attack. Though significant efforts have been made to improve the security of individual devices in these systems, little attention has been paid to security at the aggregate level. In this article, we describe a comprehensive risk management system, called GRAVITAS, for IoT/CPS that can identify undiscovered attack vectors and optimize the placement of defenses within the system for optimal performance and cost. While existing risk management systems consider only known attacks, our model employs a machine learning approach to extrapolate undiscovered exploits, enabling us to identify attacks overlooked by manual penetration testing (pen-testing). The model is flexible enough to analyze practically any IoT/CPS and provide the system administrator with a concrete list of suggested defenses that can reduce system vulnerability at optimal cost. GRAVITAS can be employed by governments, companies, and system administrators to design secure IoT/CPS at scale, providing a quantitative measure of security and efficiency in a world where IoT/CPS devices will soon be ubiquitous.

Prerit Terway, Kenza Hamidouche, Niraj K. Jha

Nonlinear system design is often a multi-objective optimization problem involving search for a design that satisfies a number of predefined constraints. The design space is typically very large since it includes all possible system architectures with different combinations of components composing each architecture. In this article, we address nonlinear system design space exploration through a two-step approach encapsulated in a framework called Fast Design Space Exploration of Nonlinear Systems (ASSENT). In the first step, we use a genetic algorithm to search for system architectures that allow discrete choices for component values or else only component values for a fixed architecture. This step yields a coarse design since the system may or may not meet the target specifications. In the second step, we use an inverse design to search over a continuous space and fine-tune the component values with the goal of improving the value of the objective function. We use a neural network to model the system response. The neural network is converted into a mixed-integer linear program for active learning to sample component values efficiently. We illustrate the efficacy of ASSENT on problems ranging from nonlinear system design to design of electrical circuits. Experimental results show that ASSENT achieves the same or better value of the objective function compared to various other optimization techniques for nonlinear system design by up to 54%. We improve sample efficiency by 6-10x compared to reinforcement learning based synthesis of electrical circuits.

Sanjai Narain, Emily Mak, Dana Chee et al.

System design tools are often only available as input-output blackboxes: for a given design as input they compute an output representing system behavior. Blackboxes are intended to be run in the forward direction. This paper presents a new method of solving the inverse design problem namely, given requirements or constraints on output, find an input that also optimizes an objective function. This problem is challenging for several reasons. First, blackboxes are not designed to be run in reverse. Second, inputs and outputs can be discrete and continuous. Third, finding designs concurrently satisfying a set of requirements is hard because designs satisfying individual requirements may conflict with each other. Fourth, blackbox evaluations can be expensive. Finally, blackboxes can sometimes fail to produce an output. This paper presents CNMA, a new method of solving the inverse problem that overcomes these challenges. CNMA tries to sample only the part of the design space relevant to solving the problem, leveraging the power of neural networks, Mixed Integer Linear Programs, and a new learning-from-failure feedback loop. The paper also presents a parallel version of CNMA that improves the efficiency and quality of solutions over the sequential version, and tries to steer it away from local optima. CNMA's performance is evaluated against conventional optimization methods for seven nonlinear design problems of 8 (two problems), 10, 15, 36 and 60 real-valued dimensions and one with 186 binary dimensions. Conventional methods evaluated are off-the-shelf implementations of Bayesian Optimization with Gaussian Processes, Nelder Mead and Random Search. The first two do not solve problems that are high-dimensional, have discrete and continuous variables or whose blackboxes can fail to return values. CNMA solves all problems, and surpasses the performance of conventional methods by up to 87%.

Malin Prematilake, Younghyun Kim, Vijay Raghunathan et al.

Implantable and wearable medical devices (IWMDs) are widely used for the monitoring and therapy of an increasing range of medical conditions. Improvements in medical devices, enabled by advances in low-power processors, more complex firmware, and wireless connectivity, have greatly improved therapeutic outcomes and patients' quality-of-life. However, security attacks, malfunctions and sometimes user errors have raised great concerns regarding the safety of IWMDs. In this work, we present a HW/SW (Hardware/Software) framework for improving the safety of IWMDs, wherein a set of safety rules and a rule check mechanism are used to monitor both the extrinsic state (the patient's physiological parameters sensed by the IWMD) and the internal state of the IWMD (I/O activities of the microcontroller) to infer unsafe operations that may be triggered by user errors, software bugs, or security attacks. We discuss how this approach can be realized in the context of a artificial pancreas with wireless connectivity and implement a prototype to demonstrate its effectiveness in improving safety at modest overheads.

Shayan Hassantabar, Joe Zhang, Hongxu Yin et al.

Mental health problems impact quality of life of millions of people around the world. However, diagnosis of mental health disorders is a challenging problem that often relies on self-reporting by patients about their behavioral patterns. Therefore, there is a need for new strategies for diagnosis of mental health problems. The recent introduction of body-area networks consisting of a plethora of accurate sensors embedded in smartwatches and smartphones and deep neural networks (DNNs) points towards a possible solution. However, disease diagnosis based on WMSs and DNNs, and their deployment on edge devices, remains a challenging problem. To this end, we propose a framework called MHDeep that utilizes commercially available WMSs and efficient DNN models to diagnose three important mental health disorders: schizoaffective, major depressive, and bipolar. MHDeep uses eight different categories of data obtained from sensors integrated in a smartwatch and smartphone. Due to limited available data, MHDeep uses a synthetic data generation module to augment real data with synthetic data drawn from the same probability distribution. We use the synthetic dataset to pre-train the DNN models, thus imposing a prior on the weights. We use a grow-and-prune DNN synthesis approach to learn both the architecture and weights during the training process. We use three different data partitions to evaluate the MHDeep models trained with data collected from 74 individuals. We conduct data instance level and patient level evaluations. MHDeep achieves an average test accuracy of 90.4%, 87.3%, and 82.4%, respectively, for classifications between healthy instances and schizoaffective disorder instances, major depressive disorder instances, and bipolar disorder instances. At the patient level, MHDeep DNNs achieve an accuracy of 100%, 100%, and 90.0% for the three mental health disorders, respectively.

Tanujay Saha, Najwa Aaraj, Neel Ajjarapu et al.

Cyber-physical systems (CPS) and Internet-of-Things (IoT) devices are increasingly being deployed across multiple functionalities, ranging from healthcare devices and wearables to critical infrastructures, e.g., nuclear power plants, autonomous vehicles, smart cities, and smart homes. These devices are inherently not secure across their comprehensive software, hardware, and network stacks, thus presenting a large attack surface that can be exploited by hackers. In this article, we present an innovative technique for detecting unknown system vulnerabilities, managing these vulnerabilities, and improving incident response when such vulnerabilities are exploited. The novelty of this approach lies in extracting intelligence from known real-world CPS/IoT attacks, representing them in the form of regular expressions, and employing machine learning (ML) techniques on this ensemble of regular expressions to generate new attack vectors and security vulnerabilities. Our results show that 10 new attack vectors and 122 new vulnerability exploits can be successfully generated that have the potential to exploit a CPS or an IoT ecosystem. The ML methodology achieves an accuracy of 97.4% and enables us to predict these attacks efficiently with an 87.2% reduction in the search space. We demonstrate the application of our method to the hacking of the in-vehicle network of a connected car. To defend against the known attacks and possible novel exploits, we discuss a defense-in-depth mechanism for various classes of attacks and the classification of data targeted by such attacks. This defense mechanism optimizes the cost of security measures based on the sensitivity of the protected resource, thus incentivizing its adoption in real-world CPS/IoT by cybersecurity practitioners.

Sanjai Narain, Emily Mak, Dana Chee et al.

Central to the design of many robot systems and their controllers is solving a constrained blackbox optimization problem. This paper presents CNMA, a new method of solving this problem that is conservative in the number of potentially expensive blackbox function evaluations; allows specifying complex, even recursive constraints directly rather than as hard-to-design penalty or barrier functions; and is resilient to the non-termination of function evaluations. CNMA leverages the ability of neural networks to approximate any continuous function, their transformation into equivalent mixed integer linear programs (MILPs) and their optimization subject to constraints with industrial strength MILP solvers. A new learning-from-failure step guides the learning to be relevant to solving the constrained optimization problem. Thus, the amount of learning is orders of magnitude smaller than that needed to learn functions over their entire domains. CNMA is illustrated with the design of several robotic systems: wave-energy propelled boat, lunar lander, hexapod, cartpole, acrobot and parallel parking. These range from 6 real-valued dimensions to 36. We show that CNMA surpasses the Nelder-Mead, Gaussian and Random Search optimization methods against the metric of number of function evaluations.

Shayan Hassantabar, Prerit Terway, Niraj K. Jha

The human brain has the ability to carry out new tasks with limited experience. It utilizes prior learning experiences to adapt the solution strategy to new domains. On the other hand, deep neural networks (DNNs) generally need large amounts of data and computational resources for training. However, this requirement is not met in many settings. To address these challenges, we propose the TUTOR DNN synthesis framework. TUTOR targets tabular datasets. It synthesizes accurate DNN models with limited available data and reduced memory/computational requirements. It consists of three sequential steps. The first step involves generation, verification, and labeling of synthetic data. The synthetic data generation module targets both the categorical and continuous features. TUTOR generates the synthetic data from the same probability distribution as the real data. It then verifies the integrity of the generated synthetic data using a semantic integrity classifier module. It labels the synthetic data based on a set of rules extracted from the real dataset. Next, TUTOR uses two training schemes that combine synthetic and training data to learn the parameters of the DNN model. These two schemes focus on two different ways in which synthetic data can be used to derive a prior on the model parameters and, hence, provide a better DNN initialization for training with real data. In the third step, TUTOR employs a grow-and-prune synthesis paradigm to learn both the weights and the architecture of the DNN to reduce model size while ensuring its accuracy. We evaluate the performance of TUTOR on nine datasets of various sizes. We show that in comparison to fully connected DNNs, TUTOR, on an average, reduces the need for data by 5.9x, improves accuracy by 3.4%, and reduces the number of parameters (fFLOPs) by 4.7x (4.3x). Thus, TUTOR enables a less data-hungry, more accurate, and more compact DNN synthesis.

Prerit Terway, Kenza Hamidouche, Niraj K. Jha

Design of cyber-physical systems (CPSs) is a challenging task that involves searching over a large search space of various CPS configurations and possible values of components composing the system. Hence, there is a need for sample-efficient CPS design space exploration to select the system architecture and component values that meet the target system requirements. We address this challenge by formulating CPS design as a multi-objective optimization problem and propose DISPATCH, a two-step methodology for sample-efficient search over the design space. First, we use a genetic algorithm to search over discrete choices of system component values for architecture search and component selection or only component selection and terminate the algorithm even before meeting the system requirements, thus yielding a coarse design. In the second step, we use an inverse design to search over a continuous space to fine-tune the component values and meet the diverse set of system requirements. We use a neural network as a surrogate function for the inverse design of the system. The neural network, converted into a mixed-integer linear program, is used for active learning to sample component values efficiently in a continuous search space. We illustrate the efficacy of DISPATCH on electrical circuit benchmarks: two-stage and three-stage transimpedence amplifiers. Simulation results show that the proposed methodology improves sample efficiency by 5-14x compared to a prior synthesis method that relies on reinforcement learning. It also synthesizes circuits with the best performance (highest bandwidth/lowest area) compared to designs synthesized using reinforcement learning, Bayesian optimization, or humans.

Wenhan Xia, Hongxu Yin, Xiaoliang Dai et al.

Modern deep neural networks are powerful and widely applicable models that extract task-relevant information through multi-level abstraction. Their cross-domain success, however, is often achieved at the expense of computational cost, high memory bandwidth, and long inference latency, which prevents their deployment in resource-constrained and time-sensitive scenarios, such as edge-side inference and self-driving cars. While recently developed methods for creating efficient deep neural networks are making their real-world deployment more feasible by reducing model size, they do not fully exploit input properties on a per-instance basis to maximize computational efficiency and task accuracy. In particular, most existing methods typically use a one-size-fits-all approach that identically processes all inputs. Motivated by the fact that different images require different feature embeddings to be accurately classified, we propose a fully dynamic paradigm that imparts deep convolutional neural networks with hierarchical inference dynamics at the level of layers and individual convolutional filters/channels. Two compact networks, called Layer-Net (L-Net) and Channel-Net (C-Net), predict on a per-instance basis which layers or filters/channels are redundant and therefore should be skipped. L-Net and C-Net also learn how to scale retained computation outputs to maximize task accuracy. By integrating L-Net and C-Net into a joint design framework, called LC-Net, we consistently outperform state-of-the-art dynamic frameworks with respect to both efficiency and classification accuracy. On the CIFAR-10 dataset, LC-Net results in up to 11.9$\times$ fewer floating-point operations (FLOPs) and up to 3.3% higher accuracy compared to other dynamic inference methods. On the ImageNet dataset, LC-Net achieves up to 1.4$\times$ fewer FLOPs and up to 4.6% higher Top-1 accuracy than the other methods.

Shayan Hassantabar, Novati Stefano, Vishweshwar Ghanakota et al.

The novel coronavirus (SARS-CoV-2) has led to a pandemic. The current testing regime based on Reverse Transcription-Polymerase Chain Reaction for SARS-CoV-2 has been unable to keep up with testing demands, and also suffers from a relatively low positive detection rate in the early stages of the resultant COVID-19 disease. Hence, there is a need for an alternative approach for repeated large-scale testing of SARS-CoV-2/COVID-19. We propose a framework called CovidDeep that combines efficient DNNs with commercially available WMSs for pervasive testing of the virus. We collected data from 87 individuals, spanning three cohorts including healthy, asymptomatic, and symptomatic patients. We trained DNNs on various subsets of the features automatically extracted from six WMS and questionnaire categories to perform ablation studies to determine which subsets are most efficacious in terms of test accuracy for a three-way classification. The highest test accuracy obtained was 98.1%. We also augmented the real training dataset with a synthetic training dataset drawn from the same probability distribution to impose a prior on DNN weights and leveraged a grow-and-prune synthesis paradigm to learn both DNN architecture and weights. This boosted the accuracy of the various DNNs further and simultaneously reduced their size and floating-point operations.

Wenhan Xia, Hongxu Yin, Niraj K. Jha

Deep neural networks (DNNs) have been deployed in myriad machine learning applications. However, advances in their accuracy are often achieved with increasingly complex and deep network architectures. These large, deep models are often unsuitable for real-world applications, due to their massive computational cost, high memory bandwidth, and long latency. For example, autonomous driving requires fast inference based on Internet-of-Things (IoT) edge devices operating under run-time energy and memory storage constraints. In such cases, compact DNNs can facilitate deployment due to their reduced energy consumption, memory requirement, and inference latency. Long short-term memories (LSTMs) are a type of recurrent neural network that have also found widespread use in the context of sequential data modeling. They also face a model size vs. accuracy trade-off. In this paper, we review major approaches for automatically synthesizing compact, yet accurate, DNN/LSTM models suitable for real-world applications. We also outline some challenges and future areas of exploration.

Shayan Hassantabar, Xiaoliang Dai, Niraj K. Jha

Neural networks (NNs) have been successfully deployed in many applications. However, architectural design of these models is still a challenging problem. Moreover, neural networks are known to have a lot of redundancy. This increases the computational cost of inference and poses an obstacle to deployment on Internet-of-Thing sensors and edge devices. To address these challenges, we propose the STEERAGE synthesis methodology. It consists of two complementary approaches: efficient architecture search, and grow-and-prune NN synthesis. The first step, covered in a global search module, uses an accuracy predictor to efficiently navigate the architectural search space. The predictor is built using boosted decision tree regression, iterative sampling, and efficient evolutionary search. The second step involves local search. By using various grow-and-prune methodologies for synthesizing convolutional and feed-forward NNs, it reduces the network redundancy, while boosting its performance. We have evaluated STEERAGE performance on various datasets, including MNIST and CIFAR-10. On MNIST dataset, our CNN architecture achieves an error rate of 0.66%, with 8.6x fewer parameters compared to the LeNet-5 baseline. For the CIFAR-10 dataset, we used the ResNet architectures as the baseline. Our STEERAGE-synthesized ResNet-18 has a 2.52% accuracy improvement over the original ResNet-18, 1.74% over ResNet-101, and 0.16% over ResNet-1001, while having comparable number of parameters and FLOPs to the original ResNet-18. This shows that instead of just increasing the number of layers to increase accuracy, an alternative is to use a better NN architecture with fewer layers. In addition, STEERAGE achieves an error rate of just 3.86% with a variant of ResNet architecture with 40 layers. To the best of our knowledge, this is the highest accuracy obtained by ResNet-based architectures on the CIFAR-10 dataset.

Hongxu Yin, Bilal Mukadam, Xiaoliang Dai et al.

Diabetes impacts the quality of life of millions of people. However, diabetes diagnosis is still an arduous process, given that the disease develops and gets treated outside the clinic. The emergence of wearable medical sensors (WMSs) and machine learning points to a way forward to address this challenge. WMSs enable a continuous mechanism to collect and analyze physiological signals. However, disease diagnosis based on WMS data and its effective deployment on resource-constrained edge devices remain challenging due to inefficient feature extraction and vast computation cost. In this work, we propose a framework called DiabDeep that combines efficient neural networks (called DiabNNs) with WMSs for pervasive diabetes diagnosis. DiabDeep bypasses the feature extraction stage and acts directly on WMS data. It enables both an (i) accurate inference on the server, e.g., a desktop, and (ii) efficient inference on an edge device, e.g., a smartphone, based on varying design goals and resource budgets. On the server, we stack sparsely connected layers to deliver high accuracy. On the edge, we use a hidden-layer long short-term memory based recurrent layer to cut down on computation and storage. At the core of DiabDeep lies a grow-and-prune training flow: it leverages gradient-based growth and magnitude-based pruning algorithms to learn both weights and connections for DiabNNs. We demonstrate the effectiveness of DiabDeep through analyzing data from 52 participants. For server (edge) side inference, we achieve a 96.3% (95.3%) accuracy in classifying diabetics against healthy individuals, and a 95.7% (94.6%) accuracy in distinguishing among type-1/type-2 diabetic, and healthy individuals. Against conventional baselines, DiabNNs achieve higher accuracy, while reducing the model size (FLOPs) by up to 454.5x (8.9x). Therefore, the system can be viewed as pervasive and efficient, yet very accurate.

Ayten Ozge Akmandor, Jorge Ortiz, Irene Manotas et al.

Supervised machine learning (ML) algorithms are aimed at maximizing classification performance under available energy and storage constraints. They try to map the training data to the corresponding labels while ensuring generalizability to unseen data. However, they do not integrate meaning-based relationships among labels in the decision process. On the other hand, natural language processing (NLP) algorithms emphasize the importance of semantic information. In this paper, we synthesize the complementary advantages of supervised ML and NLP algorithms into one method that we refer to as SECRET (Semantically Enhanced Classification of REal-world Tasks). SECRET performs classifications by fusing the semantic information of the labels with the available data: it combines the feature space of the supervised algorithms with the semantic space of the NLP algorithms and predicts labels based on this joint space. Experimental results indicate that, compared to traditional supervised learning, SECRET achieves up to 14.0% accuracy and 13.1% F1 score improvements. Moreover, compared to ensemble methods, SECRET achieves up to 12.7% accuracy and 13.3% F1 score improvements. This points to a new research direction for supervised classification based on incorporation of semantic information.

Xiaoliang Dai, Hongxu Yin, Niraj K. Jha

Deep neural networks (DNNs) have become a widely deployed model for numerous machine learning applications. However, their fixed architecture, substantial training cost, and significant model redundancy make it difficult to efficiently update them to accommodate previously unseen data. To solve these problems, we propose an incremental learning framework based on a grow-and-prune neural network synthesis paradigm. When new data arrive, the neural network first grows new connections based on the gradients to increase the network capacity to accommodate new data. Then, the framework iteratively prunes away connections based on the magnitude of weights to enhance network compactness, and hence recover efficiency. Finally, the model rests at a lightweight DNN that is both ready for inference and suitable for future grow-and-prune updates. The proposed framework improves accuracy, shrinks network size, and significantly reduces the additional training cost for incoming data compared to conventional approaches, such as training from scratch and network fine-tuning. For the LeNet-300-100 and LeNet-5 neural network architectures derived for the MNIST dataset, the framework reduces training cost by up to 64% (63%) and 67% (63%) compared to training from scratch (network fine-tuning), respectively. For the ResNet-18 architecture derived for the ImageNet dataset and DeepSpeech2 for the AN4 dataset, the corresponding training cost reductions against training from scratch (network fine-tunning) are 64% (60%) and 67% (62%), respectively. Our derived models contain fewer network parameters but achieve higher accuracy relative to conventional baselines.

Shayan Hassantabar, Zeyu Wang, Niraj K. Jha

Deep neural networks (DNNs) have become the driving force behind recent artificial intelligence (AI) research. An important problem with implementing a neural network is the design of its architecture. Typically, such an architecture is obtained manually by exploring its hyperparameter space and kept fixed during training. This approach is time-consuming and inefficient. Another issue is that modern neural networks often contain millions of parameters, whereas many applications and devices require small inference models. However, efforts to migrate DNNs to such devices typically entail a significant loss of classification accuracy. To address these challenges, we propose a two-step neural network synthesis methodology, called DR+SCANN, that combines two complementary approaches to design compact and accurate DNNs. At the core of our framework is the SCANN methodology that uses three basic architecture-changing operations, namely connection growth, neuron growth, and connection pruning, to synthesize feed-forward architectures with arbitrary structure. SCANN encapsulates three synthesis methodologies that apply a repeated grow-and-prune paradigm to three architectural starting points. DR+SCANN combines the SCANN methodology with dataset dimensionality reduction to alleviate the curse of dimensionality. We demonstrate the efficacy of SCANN and DR+SCANN on various image and non-image datasets. We evaluate SCANN on MNIST and ImageNet benchmarks. In addition, we also evaluate the efficacy of using dimensionality reduction alongside SCANN (DR+SCANN) on nine small to medium-size datasets. We also show that our synthesis methodology yields neural networks that are much better at navigating the accuracy vs. energy efficiency space. This would enable neural network-based inference even on Internet-of-Things sensors.

Hongxu Yin, Guoyang Chen, Yingmin Li et al.

Many long short-term memory (LSTM) applications need fast yet compact models. Neural network compression approaches, such as the grow-and-prune paradigm, have proved to be promising for cutting down network complexity by skipping insignificant weights. However, current compression strategies are mostly hardware-agnostic and network complexity reduction does not always translate into execution efficiency. In this work, we propose a hardware-guided symbiotic training methodology for compact, accurate, yet execution-efficient inference models. It is based on our observation that hardware may introduce substantial non-monotonic behavior, which we call the latency hysteresis effect, when evaluating network size vs. inference latency. This observation raises question about the mainstream smaller-dimension-is-better compression strategy, which often leads to a sub-optimal model architecture. By leveraging the hardware-impacted hysteresis effect and sparsity, we are able to achieve the symbiosis of model compactness and accuracy with execution efficiency, thus reducing LSTM latency while increasing its accuracy. We have evaluated our algorithms on language modeling and speech recognition applications. Relative to the traditional stacked LSTM architecture obtained for the Penn Treebank dataset, we reduce the number of parameters by 18.0x (30.5x) and measured run-time latency by up to 2.4x (5.2x) on Nvidia GPUs (Intel Xeon CPUs) without any accuracy degradation. For the DeepSpeech2 architecture obtained for the AN4 dataset, we reduce the number of parameters by 7.0x (19.4x), word error rate from 12.9% to 9.9% (10.4%), and measured run-time latency by up to 1.7x (2.4x) on Nvidia GPUs (Intel Xeon CPUs). Thus, our method yields compact, accurate, yet execution-efficient inference models.

Xiaoliang Dai, Peizhao Zhang, Bichen Wu et al.

This paper proposes an efficient neural network (NN) architecture design methodology called Chameleon that honors given resource constraints. Instead of developing new building blocks or using computationally-intensive reinforcement learning algorithms, our approach leverages existing efficient network building blocks and focuses on exploiting hardware traits and adapting computation resources to fit target latency and/or energy constraints. We formulate platform-aware NN architecture search in an optimization framework and propose a novel algorithm to search for optimal architectures aided by efficient accuracy and resource (latency and/or energy) predictors. At the core of our algorithm lies an accuracy predictor built atop Gaussian Process with Bayesian optimization for iterative sampling. With a one-time building cost for the predictors, our algorithm produces state-of-the-art model architectures on different platforms under given constraints in just minutes. Our results show that adapting computation resources to building blocks is critical to model performance. Without the addition of any bells and whistles, our models achieve significant accuracy improvements against state-of-the-art hand-crafted and automatically designed architectures. We achieve 73.8% and 75.3% top-1 accuracy on ImageNet at 20ms latency on a mobile CPU and DSP. At reduced latency, our models achieve up to 8.5% (4.8%) and 6.6% (9.3%) absolute top-1 accuracy improvements compared to MobileNetV2 and MnasNet, respectively, on a mobile CPU (DSP), and 2.7% (4.6%) and 5.6% (2.6%) accuracy gains over ResNet-101 and ResNet-152, respectively, on an Nvidia GPU (Intel CPU).

Xiaoliang Dai, Hongxu Yin, Niraj K. Jha

Long short-term memory (LSTM) has been widely used for sequential data modeling. Researchers have increased LSTM depth by stacking LSTM cells to improve performance. This incurs model redundancy, increases run-time delay, and makes the LSTMs more prone to overfitting. To address these problems, we propose a hidden-layer LSTM (H-LSTM) that adds hidden layers to LSTM's original one level non-linear control gates. H-LSTM increases accuracy while employing fewer external stacked layers, thus reducing the number of parameters and run-time latency significantly. We employ grow-and-prune (GP) training to iteratively adjust the hidden layers through gradient-based growth and magnitude-based pruning of connections. This learns both the weights and the compact architecture of H-LSTM control gates. We have GP-trained H-LSTMs for image captioning and speech recognition applications. For the NeuralTalk architecture on the MSCOCO dataset, our three models reduce the number of parameters by 38.7x [floating-point operations (FLOPs) by 45.5x], run-time latency by 4.5x, and improve the CIDEr score by 2.6. For the DeepSpeech2 architecture on the AN4 dataset, our two models reduce the number of parameters by 19.4x (FLOPs by 23.5x), run-time latency by 15.7%, and the word error rate from 12.9% to 8.7%. Thus, GP-trained H-LSTMs can be seen to be compact, fast, and accurate.